Existing Transformer-based image captioning models typically rely on the self-attention mechanism to capture long-range dependencies,which effectively extracts and leverages the global correlation of image features.Ho...Existing Transformer-based image captioning models typically rely on the self-attention mechanism to capture long-range dependencies,which effectively extracts and leverages the global correlation of image features.However,these models still face challenges in effectively capturing local associations.Moreover,since the encoder extracts global and local association features that focus on different semantic information,semantic noise may occur during the decoding stage.To address these issues,we propose the Local Relationship Enhanced Gated Transformer(LREGT).In the encoder part,we introduce the Local Relationship Enhanced Encoder(LREE),whose core component is the Local Relationship Enhanced Module(LREM).LREM consists of two novel designs:the Local Correlation Perception Module(LCPM)and the Local-Global Fusion Module(LGFM),which are beneficial for generating a comprehensive feature representation that integrates both global and local information.In the decoder part,we propose the Dual-level Multi-branch Gated Decoder(DMGD).It first creates multiple decoding branches to generate multi-perspective contextual feature representations.Subsequently,it employs the Dual-Level Gating Mechanism(DLGM)to model the multi-level relationships of these multi-perspective contextual features,enhancing their fine-grained semantics and intrinsic relationship representations.This ultimately leads to the generation of high-quality and semantically rich image captions.Experiments on the standard MSCOCO dataset demonstrate that LREGT achieves state-of-the-art performance,with a CIDEr score of 140.8 and BLEU-4 score of 41.3,significantly outperforming existing mainstream methods.These results highlight LREGT’s superiority in capturing complex visual relationships and resolving semantic noise during decoding.展开更多
Intelligence and perception are two operative technologies in 6G scenarios.The intelligent wireless network and information perception require a deep fusion of artificial intelligence(AI)and wireless communications in...Intelligence and perception are two operative technologies in 6G scenarios.The intelligent wireless network and information perception require a deep fusion of artificial intelligence(AI)and wireless communications in 6G systems.Therefore,fusion is becoming a typical feature and key challenge of 6G wireless communication systems.In this paper,we focus on the critical issues and propose three application scenarios in 6G wireless systems.Specifically,we first discuss the fusion of AI and 6G networks for the enhancement of 5G-advanced technology and future wireless communication systems.Then,we introduce the wireless AI technology architecture with 6G multidimensional information perception,which includes the physical layer technology of multi-dimensional feature information perception,full spectrum fusion technology,and intelligent wireless resource management.The discussion of key technologies for intelligent 6G wireless network networks is expected to provide a guideline for future research.展开更多
With the penetration of the Internet, virtual groups have become more and more popular. The reliability and accuracy of interpersonal perception in the virtual environment is an intriguing issue. Using the Social rela...With the penetration of the Internet, virtual groups have become more and more popular. The reliability and accuracy of interpersonal perception in the virtual environment is an intriguing issue. Using the Social relations model (SRM) [1], this paper investigates interpersonal perception in virtual groups from a multilevel perspective. In particular, it examines the following three areas: homophily, identification, and individual attraction, and explores how much of these directional and dyadic relational evaluations can be attributed to the effect of the actor, the partner, and the relationship.展开更多
Gestational Trophoblastic Disease encompasses a group of pregnancy-related disorders that derive from the placenta. Taking Leventhal’s Common Sense Model as a starting point, this study aims to investigate how illnes...Gestational Trophoblastic Disease encompasses a group of pregnancy-related disorders that derive from the placenta. Taking Leventhal’s Common Sense Model as a starting point, this study aims to investigate how illness perception could influence patients’ psychological adaptation to these rare diseases. Thirty-seven women completed: the Illness Perception Questionnaire-Revised, the Beck Depression Inventory Short Form, the State-Trait Anxiety Inventory, and the Fertility Problem Inventory. Results show that the perception of severe illness consequences significantly predicts the level of anxiety patients reported at the time of questionnaire completion. Furthermore, mental representations of illness present a significant association with infertility-related stress. Specifically, the belief in the efficacy of the treatment results in fewer feelings of discomfort and isolation from family and social context due to infertility-related problems. Since patients’ illness perception was found to have a specific impact on both anxiety and infertility-related stress, this variable should be considered in the planning of a clinical intervention.展开更多
Introduction: Lower health-related quality of life (HRQoL) is associated with fatigue, poor mental and poor gastrointestinal health during the first three months after colorectal cancer (CRC) treatment. Research indic...Introduction: Lower health-related quality of life (HRQoL) is associated with fatigue, poor mental and poor gastrointestinal health during the first three months after colorectal cancer (CRC) treatment. Research indicates that maintaining usual activities has a positive impact on HRQoL after treatment for CRC. Illness perceptions have been associated with HRQoL in other cancer diseases, and self-efficacy has been associated with HRQoL in gastrointestinal cancer survivors. Our knowledge about illness perceptions and self-efficacy in relation to maintaining everyday activities and HRQoL following CRC treatment is incomplete. Aim: To explore associations between HRQoL, fatigue, mental health, gastrointestinal health, illness perceptions and self-efficacy in relation to maintaining everyday activities, three months after surgical CRC treatment. A further aim was to test the Maintain Function Scale in a CRC population. Method: The study was cross-sectional. Forty-six persons participated. Data were collected using questionnaires. Descriptive and analytical statistics were used. Results: Persons who were more fatigued, depressed, worried, and had more diarrhea were more likely to report lower HRQoL. Increased fatigue and diarrhea were associated with decreased HRQoL. Concerning illness perceptions, persons who reported negative emotions and negative consequences of CRC were more likely to report lower HRQoL. Persons scoring higher on self-efficacy were more likely to report higher HRQoL. Increased self-efficacy was associated with increased HRQoL. The Maintain Function Scale was suitable for assessing self-efficacy in relation to maintaining everyday activities. Conclusions: Nursing support to improve self-efficacy and illness perceptions and to minimize symptoms during recovery should have a favorable impact on HRQoL.展开更多
Introduction: Many overseas volunteers have worked in low and middle-income countries. However traveling abroad and staying in an unfamiliar environment with inadequate medical infrastructure exposes volunteers to a w...Introduction: Many overseas volunteers have worked in low and middle-income countries. However traveling abroad and staying in an unfamiliar environment with inadequate medical infrastructure exposes volunteers to a wide range of health risks. The aim of this study was to clarify the relationships among disease knowledge and symptoms, disease risk perception, and self-management skills. Methods: This was a questionnaire survey of 189 Japanese who were preparing to volunteer abroad. A factor analysis was conducted on 13 items (six general risk events and seven major diseases) of risk perception. Correlation analysis was performed between the identified risk perception factors and disease knowledge and symptoms. Risk perception scores by factors from the seven disease items were compared between the Self-management Score (SMS) group and disease knowledge and risk of symptoms group, using T-tests. St. Luke’s International University Ethics Committee (19-A092) approved this study. Results: Many participants expected to suffer from digestive diseases (69.3%) and dermatitis (55%) in the field. On the other hand, few expected high risk of malaria (26.5%) or dengue fever (36%). Factor analysis yielded robust factor loadings creating a “Dread” and an “Unknown” factor accounting for 62.2% of total variance. Strong correlation was not found between SMS and disease risk perception. A weak negative correlation was observed in dengue fever, digestive disorders, and dermatitis (r = <span style="white-space:nowrap;">−0.20 to <span style="white-space:nowrap;">−0.25, p < 0.001) on the “Unknown” factor. A high SMS score was associated with a lower “Unknown” factor score for malaria, diarrhea, dermatitis, and dental disorders. In addition, higher disease knowledge was significantly associated with higher “Dread” factor score for dental disorders. Conclusions: Self-management skills and disease knowledge will enable higher risk awareness of common diseases and high lethality diseases.展开更多
<strong>Purpose:</strong> We explored parents’ perceptions and judgment formation processes concerning their infants’ health-related quality of life (HRQOL). <strong>Method:</strong> The Peds...<strong>Purpose:</strong> We explored parents’ perceptions and judgment formation processes concerning their infants’ health-related quality of life (HRQOL). <strong>Method:</strong> The PedsQL<sup>TM</sup> Infant Scales—an instrument specifically designed for infants aged 1 - 24 months—were translated into Japanese. Forward and backward translations were performed, evaluating the semantic and conceptual equivalencies. Parents with infants younger than two-years-old were recruited and interviewed using think-aloud and probing techniques. Participants completed the questionnaire while speaking aloud about what came to their mind, what they thought each question meant, and how they reached each answer. <strong>Results:</strong> Seven mothers and three fathers participated. The median age was 33.4 (28 - 43) years. Four had infants younger than six-months-old. All infants were healthy. Parents’ perceptions of their infants’ HRQOL varied across their ages. Some parents with infants younger than six months experienced difficulty discussing “emotional functioning” and “cognitive functioning” because their infants were too young to articulate the actions mentioned in the items. In those cases, the parents responded, “never a problem”. Seventy-five percent of parents recalled their infants’ daily “physical functioning”, while only 58% recalled “physical symptoms”. Some parents’ perceptions and judgment formation were compromised by their own perceptions. For example, they answered “often a problem” when the items were problematic to themselves instead of to their child. However, many distinguished their infants’ HRQOL from their own perceptions, indicating they understood the intention of the questionnaire. <strong>Conclusion:</strong> Parents’ formed judgement may compromise by their own perceptions. The result of this study will be helpful in improving healthcare communication and interpreting parents’ judgments of their infants’ HRQOL in future studies.展开更多
In intricate aquatic environments, enhancing the sensory performance of underwater actuators to ensure successful task execution is a significant challenge. To address this, a biomimetic tactile multimodal sensing app...In intricate aquatic environments, enhancing the sensory performance of underwater actuators to ensure successful task execution is a significant challenge. To address this, a biomimetic tactile multimodal sensing approach is introduced in this study, based on TriboElectric NanoGenerator (TENG) and Micro-ThermoElectric Generator (MTEG). This method enables actuators to identify the material properties of underwater target objects and to sense grasping states, such as pressure and relative sliding. In this study, a multi-dimensional underwater bionic tactile perception theoretical model is established, and a bionic sensing prototype with a sandwich-type structure is designed. To verify the performance of pressure feedback and material perception, pertinent experiments are conducted. The experimental results reveal that within a pressure measurement range of 0–16 N, the detection error of the sensor is 1.81%, and the maximum pressure response accuracy achieves 2.672 V/N. The sensing response time of the sensor is 0.981 s. The recovery time of the sensor is 0.97 s. Furthermore, the exceptional fatigue resistance of the sensor is also demonstrated. Based on the frequency of the output voltage from the prototype, the sliding state of the target object relative to the actuator can be sensed. In terms of material identification, the temperature response accuracy of the sensor is 0.072 V/°C. With the assistance of machine learning methods, six characteristic materials are identified by the sensor under 7 N pressure, with a recognition accuracy of 92.4%. In complex marine environments, this method has great application potential in the field of underwater tactile perception.展开更多
针对文本特征提取时缺乏考虑句子的上下文判别性特征以及未能充分利用实例和关系标签之间的关联信息的问题,提出一种结合语义增强和感知注意力的关系抽取方法(SPRE)。首先,在句子特征编码阶段,构建语义增强机制(SEM)提取句子的显著性语...针对文本特征提取时缺乏考虑句子的上下文判别性特征以及未能充分利用实例和关系标签之间的关联信息的问题,提出一种结合语义增强和感知注意力的关系抽取方法(SPRE)。首先,在句子特征编码阶段,构建语义增强机制(SEM)提取句子的显著性语义特征,通过实体感知词嵌入和显著特征感知(SFP)得到显著信息增强的句子表示;其次,设计感知注意力机制(PAM)整合句子特征,通过感知句子与关系标签之间的语义信息、句子的实体类型与对应关系的实体类型之间的一致性信息,以及句子之间的相似性信息评估句子与关系标签的匹配程度,充分利用包中实例与关系标签的依赖关系,进一步提高方法的降噪能力;最后,利用分类器进行关系预测并根据预测结果与实际结果的交叉熵调整网络参数。在NYT-10(New York Times 10)和GDS(Google Distant Supervision)数据集上的实验结果表明,在NYT-10数据集上,与基于BERT(Bidirectional Encoder Representations from Transformers)的关系抽取方法PARE(Passage-Attended Relation Extraction)相比,所提方法在曲线下面积(AUC)上提升了2.1个百分点,在按置信度降序排列后前100、200和300条数据精确率Precision@N(P@N)的平均值P@M上提升了2.4个百分点;在GDS数据集上,所提方法的AUC和P@M分别达到了90.5%和97.8%。所提方法在上述2个数据集上均明显优于主流的远程监督关系抽取方法,验证了该方法的有效性。可见,在主流的远程监督关系抽取任务中,所提方法能有效地提升模型对数据特征的学习能力。展开更多
基金supported by the Natural Science Foundation of China(62473105,62172118)Nature Science Key Foundation of Guangxi(2021GXNSFDA196002)+1 种基金in part by the Guangxi Key Laboratory of Image and Graphic Intelligent Processing under Grants(GIIP2302,GIIP2303,GIIP2304)Innovation Project of Guang Xi Graduate Education(2024YCXB09,2024YCXS039).
文摘Existing Transformer-based image captioning models typically rely on the self-attention mechanism to capture long-range dependencies,which effectively extracts and leverages the global correlation of image features.However,these models still face challenges in effectively capturing local associations.Moreover,since the encoder extracts global and local association features that focus on different semantic information,semantic noise may occur during the decoding stage.To address these issues,we propose the Local Relationship Enhanced Gated Transformer(LREGT).In the encoder part,we introduce the Local Relationship Enhanced Encoder(LREE),whose core component is the Local Relationship Enhanced Module(LREM).LREM consists of two novel designs:the Local Correlation Perception Module(LCPM)and the Local-Global Fusion Module(LGFM),which are beneficial for generating a comprehensive feature representation that integrates both global and local information.In the decoder part,we propose the Dual-level Multi-branch Gated Decoder(DMGD).It first creates multiple decoding branches to generate multi-perspective contextual feature representations.Subsequently,it employs the Dual-Level Gating Mechanism(DLGM)to model the multi-level relationships of these multi-perspective contextual features,enhancing their fine-grained semantics and intrinsic relationship representations.This ultimately leads to the generation of high-quality and semantically rich image captions.Experiments on the standard MSCOCO dataset demonstrate that LREGT achieves state-of-the-art performance,with a CIDEr score of 140.8 and BLEU-4 score of 41.3,significantly outperforming existing mainstream methods.These results highlight LREGT’s superiority in capturing complex visual relationships and resolving semantic noise during decoding.
文摘Intelligence and perception are two operative technologies in 6G scenarios.The intelligent wireless network and information perception require a deep fusion of artificial intelligence(AI)and wireless communications in 6G systems.Therefore,fusion is becoming a typical feature and key challenge of 6G wireless communication systems.In this paper,we focus on the critical issues and propose three application scenarios in 6G wireless systems.Specifically,we first discuss the fusion of AI and 6G networks for the enhancement of 5G-advanced technology and future wireless communication systems.Then,we introduce the wireless AI technology architecture with 6G multidimensional information perception,which includes the physical layer technology of multi-dimensional feature information perception,full spectrum fusion technology,and intelligent wireless resource management.The discussion of key technologies for intelligent 6G wireless network networks is expected to provide a guideline for future research.
文摘With the penetration of the Internet, virtual groups have become more and more popular. The reliability and accuracy of interpersonal perception in the virtual environment is an intriguing issue. Using the Social relations model (SRM) [1], this paper investigates interpersonal perception in virtual groups from a multilevel perspective. In particular, it examines the following three areas: homophily, identification, and individual attraction, and explores how much of these directional and dyadic relational evaluations can be attributed to the effect of the actor, the partner, and the relationship.
文摘Gestational Trophoblastic Disease encompasses a group of pregnancy-related disorders that derive from the placenta. Taking Leventhal’s Common Sense Model as a starting point, this study aims to investigate how illness perception could influence patients’ psychological adaptation to these rare diseases. Thirty-seven women completed: the Illness Perception Questionnaire-Revised, the Beck Depression Inventory Short Form, the State-Trait Anxiety Inventory, and the Fertility Problem Inventory. Results show that the perception of severe illness consequences significantly predicts the level of anxiety patients reported at the time of questionnaire completion. Furthermore, mental representations of illness present a significant association with infertility-related stress. Specifically, the belief in the efficacy of the treatment results in fewer feelings of discomfort and isolation from family and social context due to infertility-related problems. Since patients’ illness perception was found to have a specific impact on both anxiety and infertility-related stress, this variable should be considered in the planning of a clinical intervention.
文摘Introduction: Lower health-related quality of life (HRQoL) is associated with fatigue, poor mental and poor gastrointestinal health during the first three months after colorectal cancer (CRC) treatment. Research indicates that maintaining usual activities has a positive impact on HRQoL after treatment for CRC. Illness perceptions have been associated with HRQoL in other cancer diseases, and self-efficacy has been associated with HRQoL in gastrointestinal cancer survivors. Our knowledge about illness perceptions and self-efficacy in relation to maintaining everyday activities and HRQoL following CRC treatment is incomplete. Aim: To explore associations between HRQoL, fatigue, mental health, gastrointestinal health, illness perceptions and self-efficacy in relation to maintaining everyday activities, three months after surgical CRC treatment. A further aim was to test the Maintain Function Scale in a CRC population. Method: The study was cross-sectional. Forty-six persons participated. Data were collected using questionnaires. Descriptive and analytical statistics were used. Results: Persons who were more fatigued, depressed, worried, and had more diarrhea were more likely to report lower HRQoL. Increased fatigue and diarrhea were associated with decreased HRQoL. Concerning illness perceptions, persons who reported negative emotions and negative consequences of CRC were more likely to report lower HRQoL. Persons scoring higher on self-efficacy were more likely to report higher HRQoL. Increased self-efficacy was associated with increased HRQoL. The Maintain Function Scale was suitable for assessing self-efficacy in relation to maintaining everyday activities. Conclusions: Nursing support to improve self-efficacy and illness perceptions and to minimize symptoms during recovery should have a favorable impact on HRQoL.
文摘Introduction: Many overseas volunteers have worked in low and middle-income countries. However traveling abroad and staying in an unfamiliar environment with inadequate medical infrastructure exposes volunteers to a wide range of health risks. The aim of this study was to clarify the relationships among disease knowledge and symptoms, disease risk perception, and self-management skills. Methods: This was a questionnaire survey of 189 Japanese who were preparing to volunteer abroad. A factor analysis was conducted on 13 items (six general risk events and seven major diseases) of risk perception. Correlation analysis was performed between the identified risk perception factors and disease knowledge and symptoms. Risk perception scores by factors from the seven disease items were compared between the Self-management Score (SMS) group and disease knowledge and risk of symptoms group, using T-tests. St. Luke’s International University Ethics Committee (19-A092) approved this study. Results: Many participants expected to suffer from digestive diseases (69.3%) and dermatitis (55%) in the field. On the other hand, few expected high risk of malaria (26.5%) or dengue fever (36%). Factor analysis yielded robust factor loadings creating a “Dread” and an “Unknown” factor accounting for 62.2% of total variance. Strong correlation was not found between SMS and disease risk perception. A weak negative correlation was observed in dengue fever, digestive disorders, and dermatitis (r = <span style="white-space:nowrap;">−0.20 to <span style="white-space:nowrap;">−0.25, p < 0.001) on the “Unknown” factor. A high SMS score was associated with a lower “Unknown” factor score for malaria, diarrhea, dermatitis, and dental disorders. In addition, higher disease knowledge was significantly associated with higher “Dread” factor score for dental disorders. Conclusions: Self-management skills and disease knowledge will enable higher risk awareness of common diseases and high lethality diseases.
文摘<strong>Purpose:</strong> We explored parents’ perceptions and judgment formation processes concerning their infants’ health-related quality of life (HRQOL). <strong>Method:</strong> The PedsQL<sup>TM</sup> Infant Scales—an instrument specifically designed for infants aged 1 - 24 months—were translated into Japanese. Forward and backward translations were performed, evaluating the semantic and conceptual equivalencies. Parents with infants younger than two-years-old were recruited and interviewed using think-aloud and probing techniques. Participants completed the questionnaire while speaking aloud about what came to their mind, what they thought each question meant, and how they reached each answer. <strong>Results:</strong> Seven mothers and three fathers participated. The median age was 33.4 (28 - 43) years. Four had infants younger than six-months-old. All infants were healthy. Parents’ perceptions of their infants’ HRQOL varied across their ages. Some parents with infants younger than six months experienced difficulty discussing “emotional functioning” and “cognitive functioning” because their infants were too young to articulate the actions mentioned in the items. In those cases, the parents responded, “never a problem”. Seventy-five percent of parents recalled their infants’ daily “physical functioning”, while only 58% recalled “physical symptoms”. Some parents’ perceptions and judgment formation were compromised by their own perceptions. For example, they answered “often a problem” when the items were problematic to themselves instead of to their child. However, many distinguished their infants’ HRQOL from their own perceptions, indicating they understood the intention of the questionnaire. <strong>Conclusion:</strong> Parents’ formed judgement may compromise by their own perceptions. The result of this study will be helpful in improving healthcare communication and interpreting parents’ judgments of their infants’ HRQOL in future studies.
文摘现有的基于双向长短时记忆(BiLSTM)网络的命名实体识别(NER)模型难以全面理解文本的整体语义以及捕捉复杂的实体关系。因此,提出一种基于全域信息融合和多维关系感知的NER模型。首先,通过BERT(Bidirectional Encoder Representations from Transformers)获取输入序列的向量表示,并结合BiLSTM进一步学习输入序列的上下文信息。其次,提出由梯度稳定层和特征融合模块组成的全域信息融合机制:前者使模型保持稳定的梯度传播并更新优化输入序列的表示,后者则融合BiLSTM的前后向表示获取更全面的特征表示。接着,构建多维关系感知结构学习不同子空间单词的关联性,以捕获文档中复杂的实体关系。此外,使用自适应焦点损失函数动态调整不同类别实体的权重,提高模型对少数类实体的识别性能。最后,在7个公开数据集上将所提模型和11个基线模型进行对比,实验结果表明所提模型的F1值均优于对比模型,可见该模型的综合性较优。
基金supported by the National Natural Science Foundation of China(62372077,61976124)supported by the Fundamental Research Funds for the National Key R&D Project from the Minister of Science and Technology(2021YFA1201604).
文摘In intricate aquatic environments, enhancing the sensory performance of underwater actuators to ensure successful task execution is a significant challenge. To address this, a biomimetic tactile multimodal sensing approach is introduced in this study, based on TriboElectric NanoGenerator (TENG) and Micro-ThermoElectric Generator (MTEG). This method enables actuators to identify the material properties of underwater target objects and to sense grasping states, such as pressure and relative sliding. In this study, a multi-dimensional underwater bionic tactile perception theoretical model is established, and a bionic sensing prototype with a sandwich-type structure is designed. To verify the performance of pressure feedback and material perception, pertinent experiments are conducted. The experimental results reveal that within a pressure measurement range of 0–16 N, the detection error of the sensor is 1.81%, and the maximum pressure response accuracy achieves 2.672 V/N. The sensing response time of the sensor is 0.981 s. The recovery time of the sensor is 0.97 s. Furthermore, the exceptional fatigue resistance of the sensor is also demonstrated. Based on the frequency of the output voltage from the prototype, the sliding state of the target object relative to the actuator can be sensed. In terms of material identification, the temperature response accuracy of the sensor is 0.072 V/°C. With the assistance of machine learning methods, six characteristic materials are identified by the sensor under 7 N pressure, with a recognition accuracy of 92.4%. In complex marine environments, this method has great application potential in the field of underwater tactile perception.
文摘针对文本特征提取时缺乏考虑句子的上下文判别性特征以及未能充分利用实例和关系标签之间的关联信息的问题,提出一种结合语义增强和感知注意力的关系抽取方法(SPRE)。首先,在句子特征编码阶段,构建语义增强机制(SEM)提取句子的显著性语义特征,通过实体感知词嵌入和显著特征感知(SFP)得到显著信息增强的句子表示;其次,设计感知注意力机制(PAM)整合句子特征,通过感知句子与关系标签之间的语义信息、句子的实体类型与对应关系的实体类型之间的一致性信息,以及句子之间的相似性信息评估句子与关系标签的匹配程度,充分利用包中实例与关系标签的依赖关系,进一步提高方法的降噪能力;最后,利用分类器进行关系预测并根据预测结果与实际结果的交叉熵调整网络参数。在NYT-10(New York Times 10)和GDS(Google Distant Supervision)数据集上的实验结果表明,在NYT-10数据集上,与基于BERT(Bidirectional Encoder Representations from Transformers)的关系抽取方法PARE(Passage-Attended Relation Extraction)相比,所提方法在曲线下面积(AUC)上提升了2.1个百分点,在按置信度降序排列后前100、200和300条数据精确率Precision@N(P@N)的平均值P@M上提升了2.4个百分点;在GDS数据集上,所提方法的AUC和P@M分别达到了90.5%和97.8%。所提方法在上述2个数据集上均明显优于主流的远程监督关系抽取方法,验证了该方法的有效性。可见,在主流的远程监督关系抽取任务中,所提方法能有效地提升模型对数据特征的学习能力。
